1. Field of the Invention
The present invention generally relates to computer-implemented methods for detecting and/or sorting defects in a design pattern of a reticle. Certain embodiments relate to a computer-implemented method that includes generating a composite reference image from two or more reference images and using the composite reference image for comparison with other sample images for defect detection. Other embodiments include sorting defects using priorities, defect attributes, defect appearance and background information. Additional embodiments relate to assisting the user in locating the relevant and unique defects based on background appearance and other characteristics combined with wafer design data and knowledge of process modulation.
2. Description of the Related Art
The following description and examples are not admitted to be prior art by virtue of their inclusion in this section.
The rapid decrease in k1 (line−width=k1 (λ/NA)) in lithographic manufacture of semiconductor devices has necessitated the use of Resolution Enhancement Techniques (RET). These RET include, but are not limited to, Optical Proximity Corrections (OPC), Phase Shift Masks (PSM), and assist bar corrections. Although they are implemented in semiconductor device designs to facilitate low k1 lithography, these RET make reticles more difficult and consequently more expensive to manufacture.
Semiconductor device design and reticle manufacturing quality are verified by different procedures before the reticle enters a semiconductor fabrication facility to begin production of integrated circuits. The semiconductor device design is checked by software simulation to verify that all features print correctly after lithography in manufacturing. Such checking is commonly referred to as “Design Rule Checking.” The output of this design rule checking can produce a potentially large set of critical points, sometimes referred to as “hot spots” on the reticle layout. This set can be used to direct a point-to-point inspector, such as at a Review SEM, but this can be highly inefficient due to the number of critical points. The reticle is inspected at the mask shop for reticle defects and measured to ensure that the features are within specification. Marginal RET designs not noted by simulation checks translate into electrical failures in wafer fabrication, affect yield, and possibly remain unnoticed until wafer fabrication is complete.
Traditional methods employed in the inspection of complex mask patterns place tremendous demand on reticle inspection tools. One technique for performing image qualification entails using focus exposure matrix techniques. Performing an inspection of a conventional focus exposure matrix introduces a complication in that every exposure field is different. Die-to-die comparison is performed between adjacent local exposure fields. Any pattern change that may occur at a defocus position that is physically located farther than one exposure field from the nominal exposure field will not, therefore, be detected as different because the nominal exposure field is no longer factored in the comparison. Moreover, current reticle inspection techniques cannot detect the presence of an error in the design database. Prior art single die reticle inspection entails implementation of a design simulation technique in which a signal derived from an actual reticle is subtracted from a simulated design reference.
What is needed, therefore, is an inspection technique that is effective in locating pattern anomalies in a single die or a multi-die reticle and detecting reticle design errors resulting from errors in the design data base.
Methods have been invented to address the above-described needs. These methods are often referred to as “Process Window Qualification” Methods or “PWQ” Methods and are described in U.S. Patent Application Publication No. US2004/0091142 to Peterson et al., which is incorporated by reference as if fully set forth herein. Software packages that are configured to perform methods such as those described by Peterson et al. are commercially available from KLA-Tencor, San Jose, Calif. In general, the methods can be used to find design elements of a reticle that will fail in lithographic processing when used with lithographic variables (e.g., focus, dose, etc.) that are within a normal process window for the reticle.
PWQ methods are often performed using wafer inspection tools such as any of the wafer inspection tools that are commercially available from KLA-Tencor. In one example, a wafer is printed with columns of dies, each containing the design pattern on the reticle, in an N-M-N pattern. The “N” dies are those dies that are printed with a “nominal” lithographic variable (which may also be commonly referred to as a “nominal lithography parameter,” a “nominal lithographic process parameter,” or a “nominal process condition”). The “M” dies are printed with a value of the lithographic variable that is different than the nominal lithographic variable. In other words, the M dies are printed with a modulated lithographic variable. The nominal lithographic parameter may be the value of the lithographic parameter known to represent the “best condition” for exposure of a wafer with the reticle. Alternatively, the nominal lithographic parameter may be assigned a different baseline value of the lithographic parameter. The lithographic variable can be modulated positively and negatively with respect to the nominal lithographic variable in rows of dies printed on the wafer.
After exposure of the wafer with the reticle, the wafer is inspected by comparing the modulated die to the two nominal dies on either side of the modulated die. Adjacent dies are compared after both of the adjacent dies have been imaged. Therefore, the comparison is performed sequentially in the order in which the dies are imaged. Differences between the adjacent dies can be stored as potential defects.
Positively modulated dies and negatively modulated dies may be handled separately for purposes of analysis. In addition, the defects that are detected in the modulated dies may be analyzed to determine the priority or relevance of the defects. Furthermore, the user may be able to review the defects to find the critical or important defects that were detected.
Although the above-described PWQ methods have proved successful in meeting the needs outlined above, these methods can also be improved. For example, in the inspection process, the modulated dies are compared to exactly two nominal or reference dies. Randomly occurring defects in either or both of the reference dies may adversely affect the results if they result in reducing the priority of defects in the modulated dies. In addition, using a three die comparison (i.e., two reference dies for each modulated die) results in the use of most of the wafer area for printing the reference dies.
In the PWQ software used today, potential failure points in the design pattern are identified by looking for repeating defects. Unfortunately, by its very nature, the experiment can produce an overwhelming number of unimportant repeating defects, particularly in the dies that are highly modulated. Automatic defect classification (ADC) is one way to reduce the number of candidate defects. However, the inline ADC (iADC) method that is available for PWQ uses additional information about the defect itself, and much of this information is irrelevant to finding the most likely failure points. A newer version of the iADC method as described in U.S. patent application Ser. No. 10/954,968 to Huet et al., now U.S. Pat. No. 7,142,992 issued Nov. 28, 2006, which is incorporated by reference as if fully set forth herein, provides the capability of focusing on background features. However, in these methods, a user selects background features from the complete set of available features that are used to classify defects thereby creating an extra step in the setup of the inspection. Additionally, in current methods for reviewing defects, it is difficult to obtain multiple examples of potentially interesting defects.
The PWQ methods may also be altered to use a stored “golden die” image for comparison to the modulated images. A “golden die” image may be generally defined as an image of design pattern information on a reticle that is known in some manner to be free of defects. Therefore, by using a golden die image, the number of nominal reference dies printed on the wafer may be reduced, or even eliminated, thereby allowing more modulated dies to be printed on the wafer. However, there are disadvantages to using such a golden die image. For example, a detailed golden die image can require hundreds of Gbytes of storage. On the other hand, the detail of the golden die image may be reduced, but compromising on the detail of the golden die image compromises the effectiveness of the inspection method. Furthermore, a golden die image most likely is not formed under the same processing conditions as the test die, particularly if the golden die image is generated by simulation or if the golden die image was obtained from a different wafer than the wafer on which the modulated dies are printed. The differences in formation of the golden die and the modulated dies may result in false defect detection during inspection of the modulated dies. Moreover, reading the golden image from storage media can be slower that reacquiring the golden image from an image computer or another computer system.
Accordingly, it may be advantageous to develop computer-implemented methods for detecting and/or sorting defects in a design pattern of a reticle that allows accurate defect detection while using relatively few nominal reference dies, increases the accuracy of the defect detection by reducing the adverse effects of defects in the nominal reference dies on the accuracy of the defect detection, allows rapid identification and removal of unimportant repeating defects so that these defects do not obscure the defects of interest, allows multiple examples of interesting defects to be found relatively easily, allows classification of defects in a substantially automated manner, or achieves one or more of the above improvements without using a stored golden die image of the design pattern on the reticle.